Accuracy of open-source software segmentation and paper-based printed three-dimensional models

Piotr Szymor,Marcin Kozakiewicz,Raphael Olszewski

DOI: https://doi.org/10.1016/j.jcms.2015.11.002

Abstract:In this study, we aimed to verify the accuracy of models created with the help of open-source Slicer 3.6.3 software (Surgical Planning Lab, Harvard Medical School, Harvard University, Boston, MA, USA) and the Mcor Matrix 300 paper-based 3D printer. Our study focused on the accuracy of recreating the walls of the right orbit of a cadaveric skull. Cone beam computed tomography (CBCT) of the skull was performed (0.25-mm pixel size, 0.5-mm slice thickness). Acquired DICOM data were imported into Slicer 3.6.3 software, where segmentation was performed. A virtual model was created and saved as an .STL file and imported into Netfabb Studio professional 4.9.5 software. Three different virtual models were created by cutting the original file along three different planes (coronal, sagittal, and axial). All models were printed with a Selective Deposition Lamination Technology Matrix 300 3D printer using 80 gsm A4 paper. The models were printed so that their cutting plane was parallel to the paper sheets creating the model. Each model (coronal, sagittal, and axial) consisted of three separate parts (∼200 sheets of paper each) that were glued together to form a final model. The skull and created models were scanned with a three-dimensional (3D) optical scanner (Breuckmann smart SCAN) and were saved as .STL files. Comparisons of the orbital walls of the skull, the virtual model, and each of the three paper models were carried out with GOM Inspect 7.5SR1 software. Deviations measured between the models analysed were presented in the form of a colour-labelled map and covered with an evenly distributed network of points automatically generated by the software. An average of 804.43 ± 19.39 points for each measurement was created. Differences measured in each point were exported as a .csv file. The results were statistically analysed using Statistica 10, with statistical significance set at p < 0.05. The average number of points created on models for each measurement was 804.43 ± 19.39; however, deviation in some of the generated points could not be calculated, and those points were excluded from further calculations. From 94% to 99% of the measured absolute deviations were <1 mm. The mean absolute deviation between the skull and virtual model was 0.15 ± 0.11 mm, between the virtual and printed models was 0.15 ± 0.12 mm, and between the skull and printed models was 0.24 ± 0.21 mm. Using the optical scanner and specialized inspection software for measurements of accuracy of the created parts is recommended, as it allows one not only to measure 2-dimensional distances between anatomical points but also to perform more clinically suitable comparisons of whole surfaces. However, it requires specialized software and a very accurate scanner in order to be useful. Threshold-based, manually corrected segmentation of orbital walls performed with 3D Slicer software is accurate enough to be used for creating a virtual model of the orbit. The accuracy of the paper-based Mcor Matrix 300 3D printer is comparable to those of other commonly used 3-dimensional printers and allows one to create precise anatomical models for clinical use. The method of dividing the model into smaller parts and sticking them together seems to be quite accurate, although we recommend it only for creating small, solid models with as few parts as possible to minimize shift associated with gluing.